Download Using the transcriptome to determine the genetic mechanisms of

Common Diseases
Epistasis
Variance Heterogeneity
Using the transcriptome to determine the genetic
mechanisms of disease susceptibility
Joseph Powell
Centre for Neurogenetics and Systems Genomics
Winter School in Mathematical and Computational Biology
Brisbane - 7th July 2014
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Causes of Death - Common Diseases
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Causes of Common Diseases - Genetics + Environment
Environmental
factors
Genetics
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
How Do Mutations Influence Disease Susceptibility?
Most genetic variation underlying disease susceptibility is located
in regulatory regions
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Move To Molecular Phenotypes - Systems Genetics
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
By what means do most DNA variants alter cellular behaviour and
ultimately contribute to differences in disease susceptibility?
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Genome-wide epistasis
Joseph Powell
Variance Heterogeneity
Variance heterogeneity
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
What is epistasis?
Definition
The effect on the phenotype caused by locus A depends on the
genotype at locus B....
Epistasis has been reported in model organisms through artificial
gene knockouts, artificial line crosses and hybridization.
Our aim was to systematically search for instances of epistasis
amongst common variants for genetic variation that has arisen in
natural populations.
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Gene expression
Transcription measured for
thousands of genes
Typically heritable
Loci (eQTLs) commonly
have very large effect sizes
Good candidates to search
for epistasis
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Joseph Powell
Variance Heterogeneity
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Discovery population
846 healthy individuals
528,509 autosomal SNPs
RNA measured for
peripheral blood (Illumina
HT-12v4.0)
Expression levels for 7,339
probes
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Joseph Powell
Variance Heterogeneity
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Computational analysis
Exhaustive SNP x SNP
testing for each probe
epiGPU software and GPU
clusters
8 d.f. F-test
Over quadrillion tests
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Design Of The Analysis
Example of genotype by phenotype epistasis map
Epistatic genotype by phenotype
0.6
0.8
1.0
SNP x SNP 2 dimension analysis
0.0
0.2
0.4
RNA levels
AA
AA
Aa
Aa
SNP1
Joseph Powell
aa
Mechanisms of disease susceptibility
aa
SNP2
Common Diseases
Epistasis
Joseph Powell
Variance Heterogeneity
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Correction for multiple testing
Permutation and Bonferroni
used to give 5% FWER
Apply a family-­‐wise error rate of 5% Permutation: Single probe
FWER; T ∗ = 2.13x10−12
Correct for 7,339 probes
Experiment wide FWER;
T ∗ /7, 339 = 2.91x10−16
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Joseph Powell
Variance Heterogeneity
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Removing artifacts
For SNP pairs that passed
the 2.91x10−16 threshold
All 9 genotypes classes
present
Minimum class size of 5
No LD between SNP pairs
(r 2 < 0.1 and D 02 < 0.1)
No single loci additive or
dominance effects
11,155 pairs carried forward
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Joseph Powell
Variance Heterogeneity
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Testing for non-additive effects
Nested ANOVA for 11,155
pairs
Contrasts full genetic (8
d.f.) vs marginal effects (4
d.f)
Thus, testing for the
contribution of epistatic
variance
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Joseph Powell
Variance Heterogeneity
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
More correction for multiple testing
Epistatic effects significant
at p < 0.05/11, 155 =
4.48x10−6
501 SNP pairs with
significant interaction terms
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Joseph Powell
Variance Heterogeneity
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Replication population
Fehrmann
1,240 individuals
(Netherlands)
EGCUT
891 individuals (Estonian)
RNA measured for
peripheral blood (Illumina
HT-12v3.0)
Genotyped with Illumina
arrays
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Joseph Powell
Variance Heterogeneity
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Replication tests
Replication of 501 epistatic
hits using two independent
cohorts
Same filtering criteria
434 pairs which could be
tested in both cohorts
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Joseph Powell
Variance Heterogeneity
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Functional work
Analyses to elucidate possible
functional mechanisms
Non-synonymous mutations
GWAS and known eQTL
overlap
Genome segmentation
Chromosome interactions
Tissue specific transcription
regions
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Q-Q plots of interaction p-values from replication
150
●
The top panel shows all 434
discovery SNPs that were tested
for interactions.
100
All
●
50
●
●●
●
●
● ●
Observed
●
●
0
●
●
●
●●
●●
●●
●●●●
●●●●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
4
●
●
●
●
●
3
●
●
●
●
●
Below Bonferroni threshold
The bottom panel shows the
same data as the top panel but
excluding the 30 most significant
interactions
●
●
●
●
●
●
Matched direction of epistatic
effects p = 5.56x10−31
●
●
●
●
2
1
0
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
0
1
2
Expected
Joseph Powell
Mechanisms of disease susceptibility
3
Common Diseases
Epistasis
Variance Heterogeneity
Interactive Figure: http://kn3in.github.io/detecting_epi/
Yellow dots: A Transcript
Grey dots: Epistatic SNPs
●
●
●
● ●
●
● ●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
● ●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Joseph Powell
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
● ●
●
●
●
● ●
● ●
● ●
●
●
●
● ●
● ●
●
●
●
● ●
●
●
●
● ●
●
●
●
● ●
●
●
● ●
●
●
● ●
● ●
●
●
● ●
● ●
●
●
●
●
●
●
● ●
● ●
●
●
● ●
●
●
●
●
●
●
●
●
● ●
●
●
●
● ●
●
●
●
●
● ●
●
●
●
●
● ●
● ●
●
●
●
● ● ●
●●
●
●
● ●
●
● ● ●
●●
● ●
● ●
● ●
●
● ●
●
● ●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●●
●
●
●● ●
●
●
●
●
●
●
●
●●
●
● ● ●
● ●
● ●
●
●
● ●
●
● ●
●
● ●
●
●
●
●
●
●
● ●
● ●
●
●
●
● ●
●
●
● ●
●
●
●
●
●
● ● ●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
● ●
●
● ●
●
●
● ●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
● ●
●
●
●
●
●
●
● ●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
● ●
●
●
●
● ●
●
●
● ●
●
● ● ●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
● ●
●
●
● ●
●
●
●
●
●
●
●
● ●
●
●
●
●
● ●
●
●
● ●
● ●
●
●
● ●
● ●
●
●
●
●
●
●
●
● ●
●
● ●
●
●
● ●
●
●
●
● ●
● ●
●
●
●
●
● ●
●
●
●
● ●
● ●
●
●
●
●
●
●
●
● ●
●
●
●
●● ●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
● ●
●
●
●
● ●
●
● ●
● ●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
● ●
●
●
●
●
● ●
●
●
●
●
● ●
●
●
●
●
●
● ●
●
●
●
●
●
● ●
● ●
●
● ●
●
●
●
●
●
● ●
●
●
●
●
● ●
● ●
●
●
●
●
●
●
●
●
●
●●
●
● ●
●
● ●
●
●
●
●
●
● ●
●
●
●
●
● ●
●
●
●
● ●
●
●
●
●
● ●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
● ●
● ●
●
●
●
● ●
●
●
● ●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
● ●
●
●
●
●●
● ●
● ●
●
● ●
●
●
●
●
● ●
● ●
●
●
●● ●●
●
● ● ● ● ●
● ●
●
●
●
● ●
● ●
● ●
●
●
●● ●
● ●
● ●●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
● ●
● ●
● ●
●●
●
● ●
●
●
●
●
● ● ●●
●
●
●
●
●
●
● ● ●
● ●
●
●
●
●
●
● ● ●
●
●
●
●
● ●
●
● ●
●
● ● ●
●
● ●
●
●
●
●
●
● ● ●
●
●
●
● ●
●
● ● ●
● ●
●
●
●
● ●
●
● ●
● ●
●●
●
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Example: Epistatic Control Of CAST Regulation
CAST: Associated with numerous
rheumatic diseases
Chromosomes
It’s regulation controlled by many
epistatic SNPs throughout the
genome
Novel genetic control of regulation
potentially underlying disease
susceptibility
Joseph Powell
Epistatic
interactions
Mechanisms of disease susceptibility
Position of CAST
Gene
Common Diseases
Epistasis
Variance Heterogeneity
Functional Mechanisms - Chromosomal Interactions
750
5Mb
500
Chromosome interactions
identified in K562 cell lines
using Hi-C and ChIP-seq
0
750
1Mb
Frequency
500
250
0
750
250kb
Red lines = N SNPs within
20kb, 500kb, 2Mb and 10Mb
of known interacting regions
250
500
250
0
750
10kb
Histograms = null distributions
500
250
0
0
10
20
30
Number of chromosome interactions
Joseph Powell
Mechanisms of disease susceptibility
40
Common Diseases
Epistasis
Variance Heterogeneity
Conclusions
The first evidence for multiple instances of segregating common
polymorphisms interacting to influence human traits
Novel computational and statistical frameworks can identify and
replicate epistasis
New knowledge of the control of transcription of many genes
implicated in common disease
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Variance Heterogeneity
Searching for veQTL
There is evidence across several species for genetic control of
phenotypic variation of complex traits
This means that the variance of a phenotype is genotype dependent
Understanding genetic control of variability is important in
evolutionary biology, and medicine
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Variance eQTL
What causes them?
Epistatic interactions
Gene by Environment Interactions
Other ..... ?
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
846 healthy individuals
6,011,184 Imputed
(autosomal) SNPs
RNA measured for
peripheral blood (Illumina
HT-12v4.0)
Expression levels for 17,994
probes
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Double Generalized Linear Model
y = µ + xb + e
e ∼ N(0, σe2 )
Model to detect variance and
mean effects
e ∼ N(0, σei2 ); log (σe2 ) = c + xv
y ∼ N(µ + xb, exp(c + vx))
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
For statistical analysis, the
informational content is the same if
data is transformed by a monotonic
function
Provided the three means are
different, there is always a
monotonic transformation that will
tend to equalize the three variances
However, it’s relevance depends on
the biological scale
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Original and normalised scales
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Scale
Initial scale = Z-score from
inverse-normal transformation
Adjust original scale using transformation
functions provided in Sun et al. (2013)
2
2
σ (x) = f (m) = am + bm + c
Pictorial representation of four different non-linear
(1)
Express variance of x as a function of its mean
mx , f (mx )
Z
dx
y = g (x)∞
(2)
p
f (x)
changes to the scale. They represent four classes of
monotonic transformation (Sun et al AJHG 2013).
Re-test using dglm
The aim is to remove main driven variance effects
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Discovery results
Cis - 2.55x10−7
Trans - 3.39x10−12
747 veQTL (Potentially have
some main effects)
1412 veQTL (Potentially have
some main effects)
Only 88 have Pmain > 0.01
619 have Pmain > 0.01
Transformations remove the
effects of 655/659 (with main
effects)
Transformations remove the
effects of 732/793 (with main
effects)
Transformations remove the
effects of 1/87 (without main
effects)
Transformations remove the
effects of 21/619 (without
main effects)
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Fehrmann
1,240 individuals
(Netherlands)
EGCUT
891 individuals (Estonian)
RNA measured for
peripheral blood (Illumina
HT-12v3.0)
Genotyped with Illumina
arrays
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Replication of 7,768 veQTL
hits using two independent
cohorts
DGLM
Also Bartlett’s and Levene’s
tests
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Replication Meta-analysis
Just interested in the variance only veQTL
At FDR = 0.05
89 % of cis-veQTL
78 % of trans-veQTL
Same effect direction;
85 % cis and 77 % trans
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Ripple effect across a network
●
●
●
2.5
ILMN_1798308
●
●
0.0
−2.5
●
●
−5.0
0
1
2
factor(rs3821224)
Predicted TF binding sites tagged by expression probes. Blue lines have
variance heterogeneity with rs3821224 (p < 0.05)
Expression levels for a probe in AHSA2 by genotype class of rs3821224
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
GWAS - veSNP overlap
Disease / Trait
Asthma
Autism spectrum disorder
Bipolar disorder
Cholesterol, total
Endometriosis
HDL cholesterol
Height
Inflammatory bowel disease
LDL cholesterol
N-glycan levels
Obesity-related traits
Parkinson’s disease
Rheumatoid arthritis
Schizophrenia
Sudden cardiac arrest
Systemic lupus erythematosus
Triglycerides
Type 1 diabetes
Urate levels
Joseph Powell
N veSNPs in GWAS catalogue
1
2
2
3
1
2
4
2
6
4
5
6
3
8
2
3
7
3
2
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Conclusions
Computational and statistical frameworks can identify and
replicate variance heterogeneity for RNA transcripts.
Variance (only) eQTL represent another form of genetic control for
phenotypes.
Functional characterization of variance loci suggests a number of
possible mechanisms that underlie variance heterogeneity.
Joseph Powell
Mechanisms of disease susceptibility
Common Diseases
Epistasis
Variance Heterogeneity
Acknowledgements
Centre for Neurogenetics and Systems Genomics
QIMR Berghofer
Gib Hemani
Nick Martin
Kostya Shakhbazov
Anjali Henders
Jian Yang
Grant Montgomery
Allan Mcrae
Georgia Institute of Technology
Peter Visscher
Greg Gibson
University of Groningen
University of Tartu
Harm-Jan Westra
Andres Metspalu
Lude Franke
Tonu Esko
Dalama University
Visit us at www.complextraitgenomics.com
Lars Ronnegard
Joseph Powell
Mechanisms of disease susceptibility